TerraTransfer: Learning End-to-End Driving Policies Without Expert Demonstrations

Researchers at TerraTransfer propose a method for end-to-end autonomous driving that eliminates the need for expert demonstrations by decoupling learning to drive from learning to see. The approach uses self-play in vectorized simulators to pretrain a driving policy, then aligns its latent space with a pretrained vision backbone using action KL divergence and a structural loss. The resulting policy matches or exceeds prior end-to-end methods on photorealistic closed-loop scenarios.

arXiv:2606.17386v1 Announce Type: new Abstract: End-to-end autonomous driving has achieved state-of-the-art performance on benchmarks and real-world deployments. Its standard training recipe, however, is expensive across all stages: collecting and labeling millions of driving frames is costly, and closed-loop RL on images is bottlenecked by the per-step cost of photorealistic rendering plus a forward pass through a large vision backbone. Self-play in vectorized simulators changes the economics: millions of rollout steps per second, and a state distribution naturally rich in collisions, near-misses, and recoveries that no driving log contains. Our approach exploits this asymmetry by decoupling learning to drive from learning to see. We pretrain a single policy by self-play, then align its latent space with a pretrained vision backbone, through the action KL divergence and a batch-relational low-rank structural loss. The action target comes from the self-play policy, so alignment never supervises against a logged trajectory: a paired dataset of image, scene-state frames suffices, with no need for the curated expert demonstrations that imitation pretraining is built on. On photorealistic 3D Gaussian splatting closed-loop scenarios, the resulting end-to-end policy matches or exceeds prior end-to-end methods.